Google Launches Gemini 3.8 Live Models, Topping Speech Quality Index at Half Price

Google officially launched Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on September 16, calling them its most advanced Gemini audio models to date with SOTA performance, delivered via the Live API at frontier-model pricing. The models top Artificial Analysis' real-time speech quality index at 82.6 while halving prices, are available to try in Google AI Studio, and power the globally available Search Live experience.

Confirmed

• Per Artificial Analysis, Gemini 3.8 Live is a speech-to-speech model succeeding Gemini 3.1 Flash Live, capable of background tool and API calls during conversation, offered in a standard version and one with configurable extended thinking, at half the previous price
• Per official posts from Google AI, DeepMind and Logan Kilpatrick, the models support 97 languages with seamless mid-conversation switching, automatic language detection, and interruptible speech
• Shared features include upgraded reasoning, near-real-time visual understanding, and background (asynchronous) tool calls that don't interrupt the chat; per eherrerosj, the model can keep completing tasks after the conversation ends
• DeepMind demoed the Extended Thinking version as a coding tutor that reasons aloud without breaking the conversation
• Per philschmid, Gemini 3.8 Live tops the Artificial Analysis Quality Index at 82.6; he also shared AI Studio trial links
• Per gaganghotra, 3.8 Live now powers Search Live with more helpful answers with web links, multilingual fluency and more natural interaction, open to all users globally
• A day before launch (September 15), testingcatalog cited Bedros Pamboukian's finding that entries for both models had appeared in GCP console quota pages, confirming the pre-launch leak; koltregaskes' repost speculated a Gemini 3.8 Flash Live may follow (speculation only)
• Per a Google AI Studio announcement relayed by gaganghotra, Gemini 3.5 Transcribe was also made available alongside the new audio models via the Gemini API and AI Studio
• Per emmanuelvivier's roundup, Gemini 3.8 Live uses an audio-to-audio architecture and will roll out across Search Live, Gemini Live and Workspace, extending real-time voice beyond the assistant to search and office scenarios

Unconfirmed

• Whether and when a Gemini 3.8 Flash Live will ship has no official information, only speculation from reposts
• Pre-announcement skepticism from xiaohu, AIAndrew and koraykv about the naming and authenticity was resolved by the official launch, though details they relayed (e.g., high concurrency, low latency) lack official confirmation

Why it matters

• It continues the real-time voice agent line after last month's Gemini 3.5 Transcribe and is seen as a direct response to GPT Live (per testingcatalog)
• The combination of speaking, thinking and handling tasks in the background marks voice assistants evolving from pure conversation to task-executing real-time agents; Greg Isenberg argues over 90% of vertical SaaS will need a voice entry point (per eherrerosj)
• Topping the quality index while halving price (Artificial Analysis data) could intensify price-performance competition in real-time speech models
• Through Search Live's global rollout, the new audio capability reaches ordinary users at scale for the first time, beyond the developer API

2026-09-15 ~ 2026-09-17 · 29 related posts

• Gemini 3.8 Live and Extended Thinking variants spotted on GCP quota page, answering GPT Live 1 — testingcatalog · 2026-09-15
• Gemini 3.8 Flash Live Spotted in Google Cloud Backend, Launch Rumored Within Days — koltregaskes · 2026-09-16
• Google DeepMind launches Gemini 3.8 Live and Live Extended Thinking — GoogleDeepMind · 2026-09-16
• DeepMind demos Gemini 3.8 Live as a coding tutor that narrates its reasoning — GoogleDeepMind · 2026-09-16
• [source] Google launches Gemini 3.8 Live and 3.8 Live Extended Thinking with 97-language real-time voice — GoogleAI · 2026-09-16
• Gemini 3.8 Live launches: #1 voice model at $0.005/min with async background thinking — _philschmid · 2026-09-16
• Gemini 3.8 Live now tryable live in Google AI Studio — _philschmid · 2026-09-16
• Google's new Gemini audio models power Search Live conversations globally — gaganghotra_ · 2026-09-16
• Google Ships Gemini 3.8 Live Real-Time Voice Models with Search Live Troubleshooting — AI_Andrew · 2026-09-16
• Google Ships Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking — gibbonwalker · 2026-09-16
• Gemini 3.8 Live keeps working after the chat ends; voice front doors may redefine SaaS — eherrerosj · 2026-09-16
• Unverified: Gemini 3.8 Live voice models with reasoning announced by unofficial account — koraykv · 2026-09-16
• Google releases Gemini 3.8 Live and 3.5 Transcribe audio models for real-time voice apps — gaganghotra_ · 2026-09-16
• [source] Google releases Gemini 3.8 Live speech model, tops Speech to Speech Index at 82.6, costs $0.84/hour — ArtificialAnlys · 2026-09-16
• Google launches Gemini 3.8 Live: voice AI that chats, thinks and runs tools in parallel — 新智元 · 2026-09-16
• Google Launches Gemini 3.8 Live With Real-Time Audio-to-Audio Reasoning Across Search and Workspace — emmanuelvivier · 2026-09-16

13 near-duplicate retellings: GoogleAI · _philschmid · OfficialLoganK · OfficialLoganK · minchoi · mann_patel_6554 · leslysandra · leslysandra · xiaohu · xiaohu · heiga_zen · jocarrasqueira · eyishazyer